Changing customer demands increase the complexity and importance of production scheduling, requiring better scheduling algorithms, e.g., machine learning algorithms. At the same time, current research often neglects practical constraints, e.g., changeovers or transportation. To address this issue, we derive a representation of the scheduling problem and develop a reference architecture for future scheduling applications to increase the impact of future research. To achieve this goal, we apply a design science research approach and, first, rigorously identify the problem and derive requirements for a scheduling application based on a structured literature review. Then, we develop the problem representation and reference architecture as design science artifacts. Finally, we demonstrate the artifacts in an application scenario and publish the resulting prototypical scheduling application, enabling machine learning-based scheduling algorithms, for usage in future development projects. Our results guide future research into including practical constraints and provide practitioners with a framework for developing scheduling applications.